Improving Fairness for Data Valuation in Horizontal Federated Learning